Google releases TimesFM-3, a 330M-parameter multivariate forecasting model

Google Research has released TimesFM-3, a 330 million parameter time series foundation model that forecasts multiple related series in a single forward pass. Every TimesFM checkpoint through 2.5 was univariate: one series, its own history and nothing else.

TimesFM-3 is pretrained natively for multivariate forecasting on more than 1 trillion time points. It accepts multiple targets, past covariates and past-future covariates with no task-specific fine-tuning. Google's example is ice cream sales, where related product sales, foot traffic, weather, promotions and holidays all move the target.

Architecture

The backbone remains a decoder-only transformer. Contiguous points are grouped into patches of 32 steps and normalized per series so that wildly different scales do not dominate. Past-future covariate tokens use a lookahead trick: the current patch is concatenated with future patches, so the model sees scheduled events before they occur. Tokens then enter a 2D grid and pass through two alternating attention mechanisms.

Earlier versions decoded one patch at a time, adding latency, compute cost and compounding error. TimesFM-3 uses Contiguous Patch Masking, the training-time masking strategy introduced with TiRex. Masked placeholder tokens are appended for the whole horizon while past-future covariates stay visible, and the alternating attention layers fill every masked horizon patch simultaneously. Each target receives 9 quantiles, the 10th through 90th percentile, at every horizon step.

Results and licensing

Google evaluated the model on GIFT-Eval, fev-bench and the TIME leaderboard against Chronos-2, the Toto 2.0 family and TimesFM-2.5. According to the company, TimesFM-3 takes the top average rank among pretrained foundation models on all three, for both point and probabilistic metrics: rank 1 overall on fev-bench across 100 real-world tasks, rank 1 overall on TIME across 50 domain datasets and 98 evaluation tasks, and rank 1 among foundation models on GIFT-Eval.

Part of the repository code is Apache-2.0, but the TimesFM 3.0 weights ship under timesfm-non-commercial-license-v1.0 and are restricted to non-commercial, non-production use. You can benchmark it today; you cannot put it behind a production forecast API.

Full details are in the MarkTechPost report.